Speech Watermarking Robustness on AISHELL-3 (test)
100TPRAudioSeal
Evaluation Results
| Method | Links | |
|---|---|---|
| AudioSealAttack Category=None, Perturbation=Original, Param=-2026.06 | 100 | |
| TimbreAttack Category=None, Perturbation=Original, Param=-2026.06 | 100 | |
| WavMarkAttack Category=None, Perturbation=Original, Param=-2026.06 | 100 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=EnCodec, Param=12.0k2026.06 | 100 | |
| TimbreAttack Category=Neural Vocoders, Perturbation=BigVGAN, Param=-2026.06 | 100 | |
| TimbreAttack Category=Neural Vocoders, Perturbation=Vocos, Param=-2026.06 | 100 | |
| TimbreAttack Category=Neural Vocoders, Perturbation=HiFiGAN, Param=-2026.06 | 100 | |
| AudioSealAttack Category=Speech Enhancement, Perturbation=FRCRN, Param=-2026.06 | 100 | |
| TimbreAttack Category=Speech Enhancement, Perturbation=FRCRN, Param=-2026.06 | 100 | |
| WavMarkAttack Category=Speech Enhancement, Perturbation=FRCRN, Param=-2026.06 | 100 | |
| AudioSealAttack Category=Speech Enhancement, Perturbation=Demucs, Param=-2026.06 | 100 | |
| TimbreAttack Category=Speech Enhancement, Perturbation=Demucs, Param=-2026.06 | 100 | |
| WavMarkAttack Category=Speech Enhancement, Perturbation=Demucs, Param=-2026.06 | 100 | |
| AudioSealAttack Category=Lossy Compression, Perturbation=MP3, Param=32k2026.06 | 100 | |
| TimbreAttack Category=Lossy Compression, Perturbation=MP3, Param=32k2026.06 | 100 | |
| WavMarkAttack Category=Lossy Compression, Perturbation=MP3, Param=32k2026.06 | 100 | |
| TimbreAttack Category=Lossy Compression, Perturbation=Opus, Param=16k2026.06 | 100 | |
| AudioSealAttack Category=Lossy Compression, Perturbation=Quantization, Param=2^62026.06 | 100 | |
| TimbreAttack Category=Lossy Compression, Perturbation=Quantization, Param=2^62026.06 | 100 | |
| AudioSealAttack Category=Signal Processing, Perturbation=Gaussian Noise, Param=20dB2026.06 | 100 | |
| AudioSealAttack Category=Signal Processing, Perturbation=Background Noise, Param=20dB2026.06 | 100 | |
| TimbreAttack Category=Signal Processing, Perturbation=Background Noise, Param=20dB2026.06 | 100 | |
| AudioSealAttack Category=Signal Processing, Perturbation=Low-pass, Param=4.8kHz2026.06 | 100 | |
| TimbreAttack Category=Signal Processing, Perturbation=Low-pass, Param=4.8kHz2026.06 | 100 | |
| WavMarkAttack Category=Signal Processing, Perturbation=Low-pass, Param=4.8kHz2026.06 | 100 | |
| AudioSealAttack Category=Signal Processing, Perturbation=Smoothing, Param=182026.06 | 100 | |
| TimbreAttack Category=Signal Processing, Perturbation=Smoothing, Param=182026.06 | 100 | |
| DuraMarkAttack Category=Speech Enhancement, Perturbation=FRCRN, Param=-, Detection Protocol=Informed2026.06 | 99.9 | |
| TimbreAttack Category=Signal Processing, Perturbation=Gaussian Noise, Param=20dB2026.06 | 99.9 | |
| DuraMarkAttack Category=None, Perturbation=Original, Param=-, Detection Protocol=Informed2026.06 | 99.8 | |
| DuraMarkAttack Category=Speech Enhancement, Perturbation=Demucs, Param=-, Detection Protocol=Informed2026.06 | 99.8 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=MP3, Param=32k, Detection Protocol=Informed2026.06 | 99.8 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=BigVGAN, Param=-, Detection Protocol=Informed2026.06 | 99.7 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=Vocos, Param=-, Detection Protocol=Informed2026.06 | 99.7 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=4.5k, Detection Protocol=Informed2026.06 | 99.6 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=Quantization, Param=2^6, Detection Protocol=Informed2026.06 | 99.5 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Smoothing, Param=18, Detection Protocol=Informed2026.06 | 99.5 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=12.0k, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=3.0k, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=FACodec, Param=2.4k, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=HiFiGAN, Param=-, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=Opus, Param=16k, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Gaussian Noise, Param=20dB, Detection Protocol=Informed2026.06 | 99.4 | |
| DuraMarkAttack Category=Average, Perturbation=-, Param=-, Detection Protocol=Informed2026.06 | 99.3 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=6.0k, Detection Protocol=Informed2026.06 | 99.1 | |
| DuraMarkAttack Category=Speech Enhancement, Perturbation=FRCRN, Param=-, Detection Protocol=Blind2026.06 | 98.9 | |
| DuraMarkAttack Category=None, Perturbation=Original, Param=-, Detection Protocol=Blind2026.06 | 98.7 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=BigVGAN, Param=-, Detection Protocol=Blind2026.06 | 98.7 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Background Noise, Param=20dB, Detection Protocol=Informed2026.06 | 98.7 | |
| AudioSealAttack Category=Lossy Compression, Perturbation=Opus, Param=16k2026.06 | 98.6 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=HiFiGAN, Param=-, Detection Protocol=Blind2026.06 | 98.5 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=SpeechTokenizer, Param=4.0k, Detection Protocol=Informed2026.06 | 98.4 | |
| DuraMarkAttack Category=Speech Enhancement, Perturbation=Demucs, Param=-, Detection Protocol=Blind2026.06 | 98.3 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=MP3, Param=32k, Detection Protocol=Blind2026.06 | 98.3 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=Opus, Param=16k, Detection Protocol=Blind2026.06 | 98.3 | |
| DuraMarkAttack Category=Lossy Compression, Perturbation=Quantization, Param=2^6, Detection Protocol=Blind2026.06 | 98.2 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=3.0k, Detection Protocol=Blind2026.06 | 98 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=4.5k, Detection Protocol=Blind2026.06 | 97.9 | |
| DuraMarkAttack Category=Neural Vocoders, Perturbation=Vocos, Param=-, Detection Protocol=Blind2026.06 | 97.9 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Gaussian Noise, Param=20dB, Detection Protocol=Blind2026.06 | 97.9 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Smoothing, Param=18, Detection Protocol=Blind2026.06 | 97.8 | |
| DuraMarkAttack Category=Average, Perturbation=-, Param=-, Detection Protocol=Blind2026.06 | 97.8 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=FACodec, Param=2.4k, Detection Protocol=Blind2026.06 | 97.7 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Low-pass, Param=4.8kHz, Detection Protocol=Informed2026.06 | 97.3 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=12.0k, Detection Protocol=Blind2026.06 | 97 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=6.0k, Detection Protocol=Blind2026.06 | 96.8 | |
| DuraMarkAttack Category=Neural Codecs, Perturbation=SpeechTokenizer, Param=4.0k, Detection Protocol=Blind2026.06 | 96.6 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Background Noise, Param=20dB, Detection Protocol=Blind2026.06 | 96.5 | |
| DuraMarkAttack Category=Signal Processing, Perturbation=Low-pass, Param=4.8kHz, Detection Protocol=Blind2026.06 | 96.1 | |
| WavMarkAttack Category=Signal Processing, Perturbation=Smoothing, Param=182026.06 | 94.5 | |
| AudioSealAttack Category=Neural Vocoders, Perturbation=BigVGAN, Param=-2026.06 | 90.8 | |
| WavMarkAttack Category=Signal Processing, Perturbation=Background Noise, Param=20dB2026.06 | 90.5 | |
| TimbreAttack Category=Neural Codecs, Perturbation=DAC, Param=4.5k2026.06 | 83.2 | |
| TimbreAttack Category=Average, Perturbation=-, Param=-2026.06 | 79 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=EnCodec, Param=6.0k2026.06 | 77.3 | |
| AudioSealAttack Category=Average, Perturbation=-, Param=-2026.06 | 70.1 | |
| WavMarkAttack Category=Lossy Compression, Perturbation=Opus, Param=16k2026.06 | 66.5 | |
| TimbreAttack Category=Neural Codecs, Perturbation=EnCodec, Param=12.0k2026.06 | 55.1 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=DAC, Param=4.5k2026.06 | 51.2 | |
| TimbreAttack Category=Neural Codecs, Perturbation=DAC, Param=3.0k2026.06 | 41.4 | |
| WavMarkAttack Category=Average, Perturbation=-, Param=-2026.06 | 40.3 | |
| TimbreAttack Category=Neural Codecs, Perturbation=EnCodec, Param=6.0k2026.06 | 12.4 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=DAC, Param=3.0k2026.06 | 9.6 | |
| WavMarkAttack Category=Signal Processing, Perturbation=Gaussian Noise, Param=20dB2026.06 | 5 | |
| TimbreAttack Category=Neural Codecs, Perturbation=SpeechTokenizer, Param=4.0k2026.06 | 4.7 | |
| TimbreAttack Category=Neural Codecs, Perturbation=FACodec, Param=2.4k2026.06 | 3.6 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=SpeechTokenizer, Param=4.0k2026.06 | 1.7 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=6.0k2026.06 | 1.5 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=FACodec, Param=2.4k2026.06 | 1.3 | |
| AudioSealAttack Category=Neural Vocoders, Perturbation=HiFiGAN, Param=-2026.06 | 1.3 | |
| WavMarkAttack Category=Neural Vocoders, Perturbation=BigVGAN, Param=-2026.06 | 1.2 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=EnCodec, Param=12.0k2026.06 | 1 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=FACodec, Param=2.4k2026.06 | 1 | |
| WavMarkAttack Category=Neural Vocoders, Perturbation=Vocos, Param=-2026.06 | 1 | |
| WavMarkAttack Category=Lossy Compression, Perturbation=Quantization, Param=2^62026.06 | 1 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=3.0k2026.06 | 0.7 | |
| WavMarkAttack Category=Neural Vocoders, Perturbation=HiFiGAN, Param=-2026.06 | 0.7 | |
| AudioSealAttack Category=Neural Vocoders, Perturbation=Vocos, Param=-2026.06 | 0.5 | |
| AudioSealAttack Category=Neural Codecs, Perturbation=SpeechTokenizer, Param=4.0k2026.06 | 0.4 | |
| WavMarkAttack Category=Neural Codecs, Perturbation=DAC, Param=4.5k2026.06 | 0.2 |